Home | History | Annotate | Line # | Download | only in net
bpf.c revision 1.7
      1 /*-
      2  * Copyright (c) 1990-1991 The Regents of the University of California.
      3  * All rights reserved.
      4  *
      5  * This code is derived from the Stanford/CMU enet packet filter,
      6  * (net/enet.c) distributed as part of 4.3BSD, and code contributed
      7  * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence
      8  * Berkeley Laboratory.
      9  *
     10  * Redistribution and use in source and binary forms, with or without
     11  * modification, are permitted provided that the following conditions
     12  * are met:
     13  * 1. Redistributions of source code must retain the above copyright
     14  *    notice, this list of conditions and the following disclaimer.
     15  * 2. Redistributions in binary form must reproduce the above copyright
     16  *    notice, this list of conditions and the following disclaimer in the
     17  *    documentation and/or other materials provided with the distribution.
     18  * 3. All advertising materials mentioning features or use of this software
     19  *    must display the following acknowledgement:
     20  *	This product includes software developed by the University of
     21  *	California, Berkeley and its contributors.
     22  * 4. Neither the name of the University nor the names of its contributors
     23  *    may be used to endorse or promote products derived from this software
     24  *    without specific prior written permission.
     25  *
     26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     36  * SUCH DAMAGE.
     37  *
     38  *	from: @(#)bpf.c	7.5 (Berkeley) 7/15/91
     39  *	$Id: bpf.c,v 1.7 1993/11/23 04:51:25 cgd Exp $
     40  */
     41 
     42 #include "bpfilter.h"
     43 
     44 #if NBPFILTER > 0
     45 
     46 #include <sys/param.h>
     47 #include <sys/systm.h>
     48 #include <sys/mbuf.h>
     49 #include <sys/buf.h>
     50 #include <sys/dir.h>
     51 #include <sys/time.h>
     52 #include <sys/proc.h>
     53 #include <sys/user.h>
     54 #include <sys/ioctl.h>
     55 #include <sys/select.h>
     56 
     57 #include <sys/file.h>
     58 #if defined(sparc) && BSD < 199103
     59 #include <sys/stream.h>
     60 #endif
     61 #include <sys/tty.h>
     62 #include <sys/uio.h>
     63 
     64 #include <sys/protosw.h>
     65 #include <sys/socket.h>
     66 #include <net/if.h>
     67 
     68 #include <net/bpf.h>
     69 #include <net/bpfdesc.h>
     70 
     71 #include <sys/errno.h>
     72 
     73 #include <netinet/in.h>
     74 #include <netinet/if_ether.h>
     75 #include <sys/kernel.h>
     76 
     77 /*
     78  * Older BSDs don't have kernel malloc.
     79  */
     80 #if BSD < 199103
     81 extern bcopy();
     82 static caddr_t bpf_alloc();
     83 #include <net/bpf_compat.h>
     84 #define BPF_BUFSIZE (MCLBYTES-8)
     85 #define UIOMOVE(cp, len, code, uio) uiomove(cp, len, code, uio)
     86 #else
     87 #define BPF_BUFSIZE 4096
     88 #define UIOMOVE(cp, len, code, uio) uiomove(cp, len, uio)
     89 #endif
     90 
     91 #define PRINET  26			/* interruptible */
     92 
     93 /*
     94  * The default read buffer size is patchable.
     95  */
     96 int bpf_bufsize = BPF_BUFSIZE;
     97 
     98 /*
     99  *  bpf_iflist is the list of interfaces; each corresponds to an ifnet
    100  *  bpf_dtab holds the descriptors, indexed by minor device #
    101  */
    102 struct bpf_if	*bpf_iflist;
    103 struct bpf_d	bpf_dtab[NBPFILTER];
    104 
    105 static void	bpf_ifname();
    106 static void	catchpacket();
    107 static void	bpf_freed();
    108 static int	bpf_setif();
    109 static int	bpf_initd();
    110 static int	bpf_allocbufs();
    111 
    112 void
    113 bpfilterattach(n)
    114 	int n;
    115 {
    116 }
    117 
    118 static int
    119 bpf_movein(uio, linktype, mp, sockp)
    120 	register struct uio *uio;
    121 	int linktype;
    122 	register struct mbuf **mp;
    123 	register struct sockaddr *sockp;
    124 {
    125 	struct mbuf *m;
    126 	int error;
    127 	int len;
    128 	int hlen;
    129 
    130 	/*
    131 	 * Build a sockaddr based on the data link layer type.
    132 	 * We do this at this level because the ethernet header
    133 	 * is copied directly into the data field of the sockaddr.
    134 	 * In the case of SLIP, there is no header and the packet
    135 	 * is forwarded as is.
    136 	 * Also, we are careful to leave room at the front of the mbuf
    137 	 * for the link level header.
    138 	 */
    139 	switch (linktype) {
    140 
    141 	case DLT_SLIP:
    142 		sockp->sa_family = AF_INET;
    143 		hlen = 0;
    144 		break;
    145 
    146 	case DLT_EN10MB:
    147 		sockp->sa_family = AF_UNSPEC;
    148 		/* XXX Would MAXLINKHDR be better? */
    149 		hlen = sizeof(struct ether_header);
    150 		break;
    151 
    152 	case DLT_FDDI:
    153 		sockp->sa_family = AF_UNSPEC;
    154 		/* XXX 4(FORMAC)+6(dst)+6(src)+3(LLC)+5(SNAP) */
    155 		hlen = 24;
    156 		break;
    157 
    158 	case DLT_NULL:
    159 		sockp->sa_family = AF_UNSPEC;
    160 		hlen = 0;
    161 		break;
    162 
    163 	default:
    164 		return (EIO);
    165 	}
    166 
    167 	len = uio->uio_resid;
    168 	if ((unsigned)len > MCLBYTES)
    169 		return (EIO);
    170 
    171 	MGET(m, M_WAIT, MT_DATA);
    172 	if (m == 0)
    173 		return (ENOBUFS);
    174 	if (len > MLEN) {
    175 #if BSD >= 199103
    176 		MCLGET(m, M_WAIT);
    177 		if ((m->m_flags & M_EXT) == 0) {
    178 #else
    179 		MCLGET(m);
    180 		if (m->m_len != MCLBYTES) {
    181 #endif
    182 			error = ENOBUFS;
    183 			goto bad;
    184 		}
    185 	}
    186 	m->m_len = len;
    187 	*mp = m;
    188 	/*
    189 	 * Make room for link header.
    190 	 */
    191 	if (hlen != 0) {
    192 		m->m_len -= hlen;
    193 #if BSD >= 199103
    194 		m->m_data += hlen; /* XXX */
    195 #else
    196 		m->m_off += hlen;
    197 #endif
    198 		error = UIOMOVE((caddr_t)sockp->sa_data, hlen, UIO_WRITE, uio);
    199 		if (error)
    200 			goto bad;
    201 	}
    202 	error = UIOMOVE(mtod(m, caddr_t), len - hlen, UIO_WRITE, uio);
    203 	if (!error)
    204 		return (0);
    205  bad:
    206 	m_freem(m);
    207 	return (error);
    208 }
    209 
    210 /*
    211  * Attach file to the bpf interface, i.e. make d listen on bp.
    212  * Must be called at splimp.
    213  */
    214 static void
    215 bpf_attachd(d, bp)
    216 	struct bpf_d *d;
    217 	struct bpf_if *bp;
    218 {
    219 	/*
    220 	 * Point d at bp, and add d to the interface's list of listeners.
    221 	 * Finally, point the driver's bpf cookie at the interface so
    222 	 * it will divert packets to bpf.
    223 	 */
    224 	d->bd_bif = bp;
    225 	d->bd_next = bp->bif_dlist;
    226 	bp->bif_dlist = d;
    227 
    228 	*bp->bif_driverp = bp;
    229 }
    230 
    231 /*
    232  * Detach a file from its interface.
    233  */
    234 static void
    235 bpf_detachd(d)
    236 	struct bpf_d *d;
    237 {
    238 	struct bpf_d **p;
    239 	struct bpf_if *bp;
    240 
    241 	bp = d->bd_bif;
    242 	/*
    243 	 * Check if this descriptor had requested promiscuous mode.
    244 	 * If so, turn it off.
    245 	 */
    246 	if (d->bd_promisc) {
    247 		d->bd_promisc = 0;
    248 		if (ifpromisc(bp->bif_ifp, 0))
    249 			/*
    250 			 * Something is really wrong if we were able to put
    251 			 * the driver into promiscuous mode, but can't
    252 			 * take it out.
    253 			 */
    254 			panic("bpf: ifpromisc failed");
    255 	}
    256 	/* Remove d from the interface's descriptor list. */
    257 	p = &bp->bif_dlist;
    258 	while (*p != d) {
    259 		p = &(*p)->bd_next;
    260 		if (*p == 0)
    261 			panic("bpf_detachd: descriptor not in list");
    262 	}
    263 	*p = (*p)->bd_next;
    264 	if (bp->bif_dlist == 0)
    265 		/*
    266 		 * Let the driver know that there are no more listeners.
    267 		 */
    268 		*d->bd_bif->bif_driverp = 0;
    269 	d->bd_bif = 0;
    270 }
    271 
    272 
    273 /*
    274  * Mark a descriptor free by making it point to itself.
    275  * This is probably cheaper than marking with a constant since
    276  * the address should be in a register anyway.
    277  */
    278 #define D_ISFREE(d) ((d) == (d)->bd_next)
    279 #define D_MARKFREE(d) ((d)->bd_next = (d))
    280 #define D_MARKUSED(d) ((d)->bd_next = 0)
    281 
    282 /*
    283  * Open ethernet device.  Returns ENXIO for illegal minor device number,
    284  * EBUSY if file is open by another process.
    285  */
    286 /* ARGSUSED */
    287 int
    288 bpfopen(dev, flag)
    289 	dev_t dev;
    290 	int flag;
    291 {
    292 	register struct bpf_d *d;
    293 
    294 	if (minor(dev) >= NBPFILTER)
    295 		return (ENXIO);
    296 	/*
    297 	 * Each minor can be opened by only one process.  If the requested
    298 	 * minor is in use, return EBUSY.
    299 	 */
    300 	d = &bpf_dtab[minor(dev)];
    301 	if (!D_ISFREE(d))
    302 		return (EBUSY);
    303 
    304 	/* Mark "free" and do most initialization. */
    305 	bzero((char *)d, sizeof(*d));
    306 	d->bd_bufsize = bpf_bufsize;
    307 
    308 	return (0);
    309 }
    310 
    311 /*
    312  * Close the descriptor by detaching it from its interface,
    313  * deallocating its buffers, and marking it free.
    314  */
    315 /* ARGSUSED */
    316 int
    317 bpfclose(dev, flag)
    318 	dev_t dev;
    319 	int flag;
    320 {
    321 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
    322 	register int s;
    323 
    324 	s = splimp();
    325 	if (d->bd_bif)
    326 		bpf_detachd(d);
    327 	splx(s);
    328 	bpf_freed(d);
    329 
    330 	return (0);
    331 }
    332 
    333 /*
    334  * Support for SunOS, which does not have tsleep.
    335  */
    336 #if BSD < 199103
    337 static
    338 bpf_timeout(arg)
    339 	caddr_t arg;
    340 {
    341 	struct bpf_d *d = (struct bpf_d *)arg;
    342 	d->bd_timedout = 1;
    343 	wakeup(arg);
    344 }
    345 
    346 #define BPF_SLEEP(chan, pri, s, t) bpf_sleep((struct bpf_d *)chan)
    347 
    348 int
    349 bpf_sleep(d)
    350 	register struct bpf_d *d;
    351 {
    352 	register int rto = d->bd_rtout;
    353 	register int st;
    354 
    355 	if (rto != 0) {
    356 		d->bd_timedout = 0;
    357 		timeout(bpf_timeout, (caddr_t)d, rto);
    358 	}
    359 	st = sleep((caddr_t)d, PRINET|PCATCH);
    360 	if (rto != 0) {
    361 		if (d->bd_timedout == 0)
    362 			untimeout(bpf_timeout, (caddr_t)d);
    363 		else if (st == 0)
    364 			return EWOULDBLOCK;
    365 	}
    366 	return (st != 0) ? EINTR : 0;
    367 }
    368 #else
    369 #define BPF_SLEEP tsleep
    370 #endif
    371 
    372 /*
    373  * Rotate the packet buffers in descriptor d.  Move the store buffer
    374  * into the hold slot, and the free buffer into the store slot.
    375  * Zero the length of the new store buffer.
    376  */
    377 #define ROTATE_BUFFERS(d) \
    378 	(d)->bd_hbuf = (d)->bd_sbuf; \
    379 	(d)->bd_hlen = (d)->bd_slen; \
    380 	(d)->bd_sbuf = (d)->bd_fbuf; \
    381 	(d)->bd_slen = 0; \
    382 	(d)->bd_fbuf = 0;
    383 /*
    384  *  bpfread - read next chunk of packets from buffers
    385  */
    386 int
    387 bpfread(dev, uio)
    388 	dev_t dev;
    389 	register struct uio *uio;
    390 {
    391 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
    392 	int error;
    393 	int s;
    394 
    395 	/*
    396 	 * Restrict application to use a buffer the same size as
    397 	 * as kernel buffers.
    398 	 */
    399 	if (uio->uio_resid != d->bd_bufsize)
    400 		return (EINVAL);
    401 
    402 	s = splimp();
    403 	/*
    404 	 * If the hold buffer is empty, then do a timed sleep, which
    405 	 * ends when the timeout expires or when enough packets
    406 	 * have arrived to fill the store buffer.
    407 	 */
    408 	while (d->bd_hbuf == 0) {
    409 		if (d->bd_immediate && d->bd_slen != 0) {
    410 			/*
    411 			 * A packet(s) either arrived since the previous
    412 			 * read or arrived while we were asleep.
    413 			 * Rotate the buffers and return what's here.
    414 			 */
    415 			ROTATE_BUFFERS(d);
    416 			break;
    417 		}
    418 		error = BPF_SLEEP((caddr_t)d, PRINET|PCATCH, "bpf",
    419 				  d->bd_rtout);
    420 		if (error == EINTR || error == ERESTART) {
    421 			splx(s);
    422 			return (error);
    423 		}
    424 		if (error == EWOULDBLOCK) {
    425 			/*
    426 			 * On a timeout, return what's in the buffer,
    427 			 * which may be nothing.  If there is something
    428 			 * in the store buffer, we can rotate the buffers.
    429 			 */
    430 			if (d->bd_hbuf)
    431 				/*
    432 				 * We filled up the buffer in between
    433 				 * getting the timeout and arriving
    434 				 * here, so we don't need to rotate.
    435 				 */
    436 				break;
    437 
    438 			if (d->bd_slen == 0) {
    439 				splx(s);
    440 				return (0);
    441 			}
    442 			ROTATE_BUFFERS(d);
    443 			break;
    444 		}
    445 	}
    446 	/*
    447 	 * At this point, we know we have something in the hold slot.
    448 	 */
    449 	splx(s);
    450 
    451 	/*
    452 	 * Move data from hold buffer into user space.
    453 	 * We know the entire buffer is transferred since
    454 	 * we checked above that the read buffer is bpf_bufsize bytes.
    455 	 */
    456 	error = UIOMOVE(d->bd_hbuf, d->bd_hlen, UIO_READ, uio);
    457 
    458 	s = splimp();
    459 	d->bd_fbuf = d->bd_hbuf;
    460 	d->bd_hbuf = 0;
    461 	d->bd_hlen = 0;
    462 	splx(s);
    463 
    464 	return (error);
    465 }
    466 
    467 
    468 /*
    469  * If there are processes sleeping on this descriptor, wake them up.
    470  */
    471 static inline void
    472 bpf_wakeup(d)
    473 	register struct bpf_d *d;
    474 {
    475 	wakeup((caddr_t)d);
    476 #if (BSD > 199103) || defined(__NetBSD__)
    477 	selwakeup(&d->bd_sel);
    478 	/* XXX */
    479 	d->bd_sel.si_pid = 0;
    480 #else
    481 	if (d->bd_selproc) {
    482 		selwakeup(d->bd_selproc, (int)d->bd_selcoll);
    483 		d->bd_selcoll = 0;
    484 		d->bd_selproc = 0;
    485 	}
    486 #endif
    487 }
    488 
    489 int
    490 bpfwrite(dev, uio)
    491 	dev_t dev;
    492 	struct uio *uio;
    493 {
    494 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
    495 	struct ifnet *ifp;
    496 	struct mbuf *m;
    497 	int error, s;
    498 	static struct sockaddr dst;
    499 
    500 	if (d->bd_bif == 0)
    501 		return (ENXIO);
    502 
    503 	ifp = d->bd_bif->bif_ifp;
    504 
    505 	if (uio->uio_resid == 0)
    506 		return (0);
    507 	if (uio->uio_resid > ifp->if_mtu)
    508 		return (EMSGSIZE);
    509 
    510 	error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, &m, &dst);
    511 	if (error)
    512 		return (error);
    513 
    514 	s = splnet();
    515 #if BSD >= 199103
    516 	error = (*ifp->if_output)(ifp, m, &dst, (struct rtentry *)0);
    517 #else
    518 	error = (*ifp->if_output)(ifp, m, &dst);
    519 #endif
    520 	splx(s);
    521 	/*
    522 	 * The driver frees the mbuf.
    523 	 */
    524 	return (error);
    525 }
    526 
    527 /*
    528  * Reset a descriptor by flushing its packet buffer and clearing the
    529  * receive and drop counts.  Should be called at splimp.
    530  */
    531 static void
    532 reset_d(d)
    533 	struct bpf_d *d;
    534 {
    535 	if (d->bd_hbuf) {
    536 		/* Free the hold buffer. */
    537 		d->bd_fbuf = d->bd_hbuf;
    538 		d->bd_hbuf = 0;
    539 	}
    540 	d->bd_slen = 0;
    541 	d->bd_hlen = 0;
    542 	d->bd_rcount = 0;
    543 	d->bd_dcount = 0;
    544 }
    545 
    546 /*
    547  *  FIONREAD		Check for read packet available.
    548  *  SIOCGIFADDR		Get interface address - convenient hook to driver.
    549  *  BIOCGBLEN		Get buffer len [for read()].
    550  *  BIOCSETF		Set ethernet read filter.
    551  *  BIOCFLUSH		Flush read packet buffer.
    552  *  BIOCPROMISC		Put interface into promiscuous mode.
    553  *  BIOCGDLT		Get link layer type.
    554  *  BIOCGETIF		Get interface name.
    555  *  BIOCSETIF		Set interface.
    556  *  BIOCSRTIMEOUT	Set read timeout.
    557  *  BIOCGRTIMEOUT	Get read timeout.
    558  *  BIOCGSTATS		Get packet stats.
    559  *  BIOCIMMEDIATE	Set immediate mode.
    560  *  BIOCVERSION		Get filter language version.
    561  */
    562 /* ARGSUSED */
    563 int
    564 bpfioctl(dev, cmd, addr, flag)
    565 	dev_t dev;
    566 	int cmd;
    567 	caddr_t addr;
    568 	int flag;
    569 {
    570 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
    571 	int s, error = 0;
    572 
    573 	switch (cmd) {
    574 
    575 	default:
    576 		error = EINVAL;
    577 		break;
    578 
    579 	/*
    580 	 * Check for read packet available.
    581 	 */
    582 	case FIONREAD:
    583 		{
    584 			int n;
    585 
    586 			s = splimp();
    587 			n = d->bd_slen;
    588 			if (d->bd_hbuf)
    589 				n += d->bd_hlen;
    590 			splx(s);
    591 
    592 			*(int *)addr = n;
    593 			break;
    594 		}
    595 
    596 	case SIOCGIFADDR:
    597 		{
    598 			struct ifnet *ifp;
    599 
    600 			if (d->bd_bif == 0)
    601 				error = EINVAL;
    602 			else {
    603 				ifp = d->bd_bif->bif_ifp;
    604 				error = (*ifp->if_ioctl)(ifp, cmd, addr);
    605 			}
    606 			break;
    607 		}
    608 
    609 	/*
    610 	 * Get buffer len [for read()].
    611 	 */
    612 	case BIOCGBLEN:
    613 		*(u_int *)addr = d->bd_bufsize;
    614 		break;
    615 
    616 	/*
    617 	 * Set buffer length.
    618 	 */
    619 	case BIOCSBLEN:
    620 #if BSD < 199103
    621 		error = EINVAL;
    622 #else
    623 		if (d->bd_bif != 0)
    624 			error = EINVAL;
    625 		else {
    626 			register u_int size = *(u_int *)addr;
    627 
    628 			if (size > BPF_MAXBUFSIZE)
    629 				*(u_int *)addr = size = BPF_MAXBUFSIZE;
    630 			else if (size < BPF_MINBUFSIZE)
    631 				*(u_int *)addr = size = BPF_MINBUFSIZE;
    632 			d->bd_bufsize = size;
    633 		}
    634 #endif
    635 		break;
    636 
    637 	/*
    638 	 * Set link layer read filter.
    639 	 */
    640 	case BIOCSETF:
    641 		error = bpf_setf(d, (struct bpf_program *)addr);
    642 		break;
    643 
    644 	/*
    645 	 * Flush read packet buffer.
    646 	 */
    647 	case BIOCFLUSH:
    648 		s = splimp();
    649 		reset_d(d);
    650 		splx(s);
    651 		break;
    652 
    653 	/*
    654 	 * Put interface into promiscuous mode.
    655 	 */
    656 	case BIOCPROMISC:
    657 		if (d->bd_bif == 0) {
    658 			/*
    659 			 * No interface attached yet.
    660 			 */
    661 			error = EINVAL;
    662 			break;
    663 		}
    664 		s = splimp();
    665 		if (d->bd_promisc == 0) {
    666 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
    667 			if (error == 0)
    668 				d->bd_promisc = 1;
    669 		}
    670 		splx(s);
    671 		break;
    672 
    673 	/*
    674 	 * Get device parameters.
    675 	 */
    676 	case BIOCGDLT:
    677 		if (d->bd_bif == 0)
    678 			error = EINVAL;
    679 		else
    680 			*(u_int *)addr = d->bd_bif->bif_dlt;
    681 		break;
    682 
    683 	/*
    684 	 * Set interface name.
    685 	 */
    686 	case BIOCGETIF:
    687 		if (d->bd_bif == 0)
    688 			error = EINVAL;
    689 		else
    690 			bpf_ifname(d->bd_bif->bif_ifp, (struct ifreq *)addr);
    691 		break;
    692 
    693 	/*
    694 	 * Set interface.
    695 	 */
    696 	case BIOCSETIF:
    697 		error = bpf_setif(d, (struct ifreq *)addr);
    698 		break;
    699 
    700 	/*
    701 	 * Set read timeout.
    702 	 */
    703 	case BIOCSRTIMEOUT:
    704 		{
    705 			struct timeval *tv = (struct timeval *)addr;
    706 			u_long msec;
    707 
    708 			/* Compute number of milliseconds. */
    709 			msec = tv->tv_sec * 1000 + tv->tv_usec / 1000;
    710 			/* Scale milliseconds to ticks.  Assume hard
    711 			   clock has millisecond or greater resolution
    712 			   (i.e. tick >= 1000).  For 10ms hardclock,
    713 			   tick/1000 = 10, so rtout<-msec/10. */
    714 			d->bd_rtout = msec / (tick / 1000);
    715 			break;
    716 		}
    717 
    718 	/*
    719 	 * Get read timeout.
    720 	 */
    721 	case BIOCGRTIMEOUT:
    722 		{
    723 			struct timeval *tv = (struct timeval *)addr;
    724 			u_long msec = d->bd_rtout;
    725 
    726 			msec *= tick / 1000;
    727 			tv->tv_sec = msec / 1000;
    728 			tv->tv_usec = msec % 1000;
    729 			break;
    730 		}
    731 
    732 	/*
    733 	 * Get packet stats.
    734 	 */
    735 	case BIOCGSTATS:
    736 		{
    737 			struct bpf_stat *bs = (struct bpf_stat *)addr;
    738 
    739 			bs->bs_recv = d->bd_rcount;
    740 			bs->bs_drop = d->bd_dcount;
    741 			break;
    742 		}
    743 
    744 	/*
    745 	 * Set immediate mode.
    746 	 */
    747 	case BIOCIMMEDIATE:
    748 		d->bd_immediate = *(u_int *)addr;
    749 		break;
    750 
    751 	case BIOCVERSION:
    752 		{
    753 			struct bpf_version *bv = (struct bpf_version *)addr;
    754 
    755 			bv->bv_major = BPF_MAJOR_VERSION;
    756 			bv->bv_minor = BPF_MINOR_VERSION;
    757 			break;
    758 		}
    759 	}
    760 	return (error);
    761 }
    762 
    763 /*
    764  * Set d's packet filter program to fp.  If this file already has a filter,
    765  * free it and replace it.  Returns EINVAL for bogus requests.
    766  */
    767 int
    768 bpf_setf(d, fp)
    769 	struct bpf_d *d;
    770 	struct bpf_program *fp;
    771 {
    772 	struct bpf_insn *fcode, *old;
    773 	u_int flen, size;
    774 	int s;
    775 
    776 	old = d->bd_filter;
    777 	if (fp->bf_insns == 0) {
    778 		if (fp->bf_len != 0)
    779 			return (EINVAL);
    780 		s = splimp();
    781 		d->bd_filter = 0;
    782 		reset_d(d);
    783 		splx(s);
    784 		if (old != 0)
    785 			free((caddr_t)old, M_DEVBUF);
    786 		return (0);
    787 	}
    788 	flen = fp->bf_len;
    789 	if (flen > BPF_MAXINSNS)
    790 		return (EINVAL);
    791 
    792 	size = flen * sizeof(*fp->bf_insns);
    793 	fcode = (struct bpf_insn *)malloc(size, M_DEVBUF, M_WAITOK);
    794 	if (copyin((caddr_t)fp->bf_insns, (caddr_t)fcode, size) == 0 &&
    795 	    bpf_validate(fcode, (int)flen)) {
    796 		s = splimp();
    797 		d->bd_filter = fcode;
    798 		reset_d(d);
    799 		splx(s);
    800 		if (old != 0)
    801 			free((caddr_t)old, M_DEVBUF);
    802 
    803 		return (0);
    804 	}
    805 	free((caddr_t)fcode, M_DEVBUF);
    806 	return (EINVAL);
    807 }
    808 
    809 /*
    810  * Detach a file from its current interface (if attached at all) and attach
    811  * to the interface indicated by the name stored in ifr.
    812  * Return an errno or 0.
    813  */
    814 static int
    815 bpf_setif(d, ifr)
    816 	struct bpf_d *d;
    817 	struct ifreq *ifr;
    818 {
    819 	struct bpf_if *bp;
    820 	char *cp;
    821 	int unit, s, error;
    822 
    823 	/*
    824 	 * Separate string into name part and unit number.  Put a null
    825 	 * byte at the end of the name part, and compute the number.
    826 	 * If the a unit number is unspecified, the default is 0,
    827 	 * as initialized above.  XXX This should be common code.
    828 	 */
    829 	unit = 0;
    830 	cp = ifr->ifr_name;
    831 	cp[sizeof(ifr->ifr_name) - 1] = '\0';
    832 	while (*cp++) {
    833 		if (*cp >= '0' && *cp <= '9') {
    834 			unit = *cp - '0';
    835 			*cp++ = '\0';
    836 			while (*cp)
    837 				unit = 10 * unit + *cp++ - '0';
    838 			break;
    839 		}
    840 	}
    841 	/*
    842 	 * Look through attached interfaces for the named one.
    843 	 */
    844 	for (bp = bpf_iflist; bp != 0; bp = bp->bif_next) {
    845 		struct ifnet *ifp = bp->bif_ifp;
    846 
    847 		if (ifp == 0 || unit != ifp->if_unit
    848 		    || strcmp(ifp->if_name, ifr->ifr_name) != 0)
    849 			continue;
    850 		/*
    851 		 * We found the requested interface.
    852 		 * If it's not up, return an error.
    853 		 * Allocate the packet buffers if we need to.
    854 		 * If we're already attached to requested interface,
    855 		 * just flush the buffer.
    856 		 */
    857 		if ((ifp->if_flags & IFF_UP) == 0)
    858 			return (ENETDOWN);
    859 
    860 		if (d->bd_sbuf == 0) {
    861 			error = bpf_allocbufs(d);
    862 			if (error != 0)
    863 				return (error);
    864 		}
    865 		s = splimp();
    866 		if (bp != d->bd_bif) {
    867 			if (d->bd_bif)
    868 				/*
    869 				 * Detach if attached to something else.
    870 				 */
    871 				bpf_detachd(d);
    872 
    873 			bpf_attachd(d, bp);
    874 		}
    875 		reset_d(d);
    876 		splx(s);
    877 		return (0);
    878 	}
    879 	/* Not found. */
    880 	return (ENXIO);
    881 }
    882 
    883 /*
    884  * Convert an interface name plus unit number of an ifp to a single
    885  * name which is returned in the ifr.
    886  */
    887 static void
    888 bpf_ifname(ifp, ifr)
    889 	struct ifnet *ifp;
    890 	struct ifreq *ifr;
    891 {
    892 	char *s = ifp->if_name;
    893 	char *d = ifr->ifr_name;
    894 
    895 	while (*d++ = *s++)
    896 		continue;
    897 	/* XXX Assume that unit number is less than 10. */
    898 	*d++ = ifp->if_unit + '0';
    899 	*d = '\0';
    900 }
    901 
    902 /*
    903  * The new select interface passes down the proc pointer; the old select
    904  * stubs had to grab it out of the user struct.  This glue allows either case.
    905  */
    906 #if BSD >= 199103
    907 #define bpf_select bpfselect
    908 #else
    909 int
    910 bpfselect(dev, rw)
    911 	register dev_t dev;
    912 	int rw;
    913 {
    914 	return (bpf_select(dev, rw, u.u_procp));
    915 }
    916 #endif
    917 
    918 /*
    919  * Support for select() system call
    920  * Inspired by the code in tty.c for the same purpose.
    921  *
    922  * Return true iff the specific operation will not block indefinitely.
    923  * Otherwise, return false but make a note that a selwakeup() must be done.
    924  */
    925 int
    926 bpf_select(dev, rw, p)
    927 	register dev_t dev;
    928 	int rw;
    929 	struct proc *p;
    930 {
    931 	register struct bpf_d *d;
    932 	register int s;
    933 
    934 	if (rw != FREAD)
    935 		return (0);
    936 	/*
    937 	 * An imitation of the FIONREAD ioctl code.
    938 	 */
    939 	d = &bpf_dtab[minor(dev)];
    940 
    941 	s = splimp();
    942 	if (d->bd_hlen != 0 || (d->bd_immediate && d->bd_slen != 0)) {
    943 		/*
    944 		 * There is data waiting.
    945 		 */
    946 		splx(s);
    947 		return (1);
    948 	}
    949 #if defined(__NetBSD__)
    950 	selrecord(p, &d->bd_sel);
    951 #else
    952 	/*
    953 	 * No data ready.  If there's already a select() waiting on this
    954 	 * minor device then this is a collision.  This shouldn't happen
    955 	 * because minors really should not be shared, but if a process
    956 	 * forks while one of these is open, it is possible that both
    957 	 * processes could select on the same descriptor.
    958 	 */
    959 	if (d->bd_selproc && d->bd_selproc->p_wchan == (caddr_t)&selwait)
    960 		d->bd_selcoll = 1;
    961 	else
    962 		d->bd_selproc = p;
    963 #endif
    964 	splx(s);
    965 	return (0);
    966 }
    967 
    968 /*
    969  * Incoming linkage from device drivers.  Process the packet pkt, of length
    970  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
    971  * by each process' filter, and if accepted, stashed into the corresponding
    972  * buffer.
    973  */
    974 void
    975 bpf_tap(arg, pkt, pktlen)
    976 	caddr_t arg;
    977 	register u_char *pkt;
    978 	register u_int pktlen;
    979 {
    980 	struct bpf_if *bp;
    981 	register struct bpf_d *d;
    982 	register u_int slen;
    983 	/*
    984 	 * Note that the ipl does not have to be raised at this point.
    985 	 * The only problem that could arise here is that if two different
    986 	 * interfaces shared any data.  This is not the case.
    987 	 */
    988 	bp = (struct bpf_if *)arg;
    989 	for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
    990 		++d->bd_rcount;
    991 		slen = bpf_filter(d->bd_filter, pkt, pktlen, pktlen);
    992 		if (slen != 0)
    993 			catchpacket(d, pkt, pktlen, slen, bcopy);
    994 	}
    995 }
    996 
    997 /*
    998  * Copy data from an mbuf chain into a buffer.  This code is derived
    999  * from m_copydata in sys/uipc_mbuf.c.
   1000  */
   1001 static void
   1002 bpf_mcopy(src, dst, len)
   1003 	u_char *src;
   1004 	u_char *dst;
   1005 	register int len;
   1006 {
   1007 	register struct mbuf *m = (struct mbuf *)src;
   1008 	register unsigned count;
   1009 
   1010 	while (len > 0) {
   1011 		if (m == 0)
   1012 			panic("bpf_mcopy");
   1013 		count = MIN(m->m_len, len);
   1014 		bcopy(mtod(m, caddr_t), (caddr_t)dst, count);
   1015 		m = m->m_next;
   1016 		dst += count;
   1017 		len -= count;
   1018 	}
   1019 }
   1020 
   1021 /*
   1022  * Incoming linkage from device drivers, when packet is in an mbuf chain.
   1023  */
   1024 void
   1025 bpf_mtap(arg, m)
   1026 	caddr_t arg;
   1027 	struct mbuf *m;
   1028 {
   1029 	struct bpf_if *bp = (struct bpf_if *)arg;
   1030 	struct bpf_d *d;
   1031 	u_int pktlen, slen;
   1032 	struct mbuf *m0;
   1033 
   1034 	pktlen = 0;
   1035 	for (m0 = m; m0 != 0; m0 = m0->m_next)
   1036 		pktlen += m0->m_len;
   1037 
   1038 	for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
   1039 		++d->bd_rcount;
   1040 		slen = bpf_filter(d->bd_filter, (u_char *)m, pktlen, 0);
   1041 		if (slen != 0)
   1042 			catchpacket(d, (u_char *)m, pktlen, slen, bpf_mcopy);
   1043 	}
   1044 }
   1045 
   1046 /*
   1047  * Move the packet data from interface memory (pkt) into the
   1048  * store buffer.  Return 1 if it's time to wakeup a listener (buffer full),
   1049  * otherwise 0.  "copy" is the routine called to do the actual data
   1050  * transfer.  bcopy is passed in to copy contiguous chunks, while
   1051  * bpf_mcopy is passed in to copy mbuf chains.  In the latter case,
   1052  * pkt is really an mbuf.
   1053  */
   1054 static void
   1055 catchpacket(d, pkt, pktlen, snaplen, cpfn)
   1056 	register struct bpf_d *d;
   1057 	register u_char *pkt;
   1058 	register u_int pktlen, snaplen;
   1059 	register void (*cpfn)();
   1060 {
   1061 	register struct bpf_hdr *hp;
   1062 	register int totlen, curlen;
   1063 	register int hdrlen = d->bd_bif->bif_hdrlen;
   1064 	/*
   1065 	 * Figure out how many bytes to move.  If the packet is
   1066 	 * greater or equal to the snapshot length, transfer that
   1067 	 * much.  Otherwise, transfer the whole packet (unless
   1068 	 * we hit the buffer size limit).
   1069 	 */
   1070 	totlen = hdrlen + MIN(snaplen, pktlen);
   1071 	if (totlen > d->bd_bufsize)
   1072 		totlen = d->bd_bufsize;
   1073 
   1074 	/*
   1075 	 * Round up the end of the previous packet to the next longword.
   1076 	 */
   1077 	curlen = BPF_WORDALIGN(d->bd_slen);
   1078 	if (curlen + totlen > d->bd_bufsize) {
   1079 		/*
   1080 		 * This packet will overflow the storage buffer.
   1081 		 * Rotate the buffers if we can, then wakeup any
   1082 		 * pending reads.
   1083 		 */
   1084 		if (d->bd_fbuf == 0) {
   1085 			/*
   1086 			 * We haven't completed the previous read yet,
   1087 			 * so drop the packet.
   1088 			 */
   1089 			++d->bd_dcount;
   1090 			return;
   1091 		}
   1092 		ROTATE_BUFFERS(d);
   1093 		bpf_wakeup(d);
   1094 		curlen = 0;
   1095 	}
   1096 	else if (d->bd_immediate)
   1097 		/*
   1098 		 * Immediate mode is set.  A packet arrived so any
   1099 		 * reads should be woken up.
   1100 		 */
   1101 		bpf_wakeup(d);
   1102 
   1103 	/*
   1104 	 * Append the bpf header.
   1105 	 */
   1106 	hp = (struct bpf_hdr *)(d->bd_sbuf + curlen);
   1107 #if BSD >= 199103
   1108 	microtime(&hp->bh_tstamp);
   1109 #elif defined(sun)
   1110 	uniqtime(&hp->bh_tstamp);
   1111 #else
   1112 	hp->bh_tstamp = time;
   1113 #endif
   1114 	hp->bh_datalen = pktlen;
   1115 	hp->bh_hdrlen = hdrlen;
   1116 	/*
   1117 	 * Copy the packet data into the store buffer and update its length.
   1118 	 */
   1119 	(*cpfn)(pkt, (u_char *)hp + hdrlen, (hp->bh_caplen = totlen - hdrlen));
   1120 	d->bd_slen = curlen + totlen;
   1121 }
   1122 
   1123 /*
   1124  * Initialize all nonzero fields of a descriptor.
   1125  */
   1126 static int
   1127 bpf_allocbufs(d)
   1128 	register struct bpf_d *d;
   1129 {
   1130 	d->bd_fbuf = (caddr_t)malloc(d->bd_bufsize, M_DEVBUF, M_WAITOK);
   1131 	if (d->bd_fbuf == 0)
   1132 		return (ENOBUFS);
   1133 
   1134 	d->bd_sbuf = (caddr_t)malloc(d->bd_bufsize, M_DEVBUF, M_WAITOK);
   1135 	if (d->bd_sbuf == 0) {
   1136 		free(d->bd_fbuf, M_DEVBUF);
   1137 		return (ENOBUFS);
   1138 	}
   1139 	d->bd_slen = 0;
   1140 	d->bd_hlen = 0;
   1141 	return (0);
   1142 }
   1143 
   1144 /*
   1145  * Free buffers currently in use by a descriptor.
   1146  * Called on close.
   1147  */
   1148 static void
   1149 bpf_freed(d)
   1150 	register struct bpf_d *d;
   1151 {
   1152 	/*
   1153 	 * We don't need to lock out interrupts since this descriptor has
   1154 	 * been detached from its interface and it yet hasn't been marked
   1155 	 * free.
   1156 	 */
   1157 	if (d->bd_sbuf != 0) {
   1158 		free(d->bd_sbuf, M_DEVBUF);
   1159 		if (d->bd_hbuf != 0)
   1160 			free(d->bd_hbuf, M_DEVBUF);
   1161 		if (d->bd_fbuf != 0)
   1162 			free(d->bd_fbuf, M_DEVBUF);
   1163 	}
   1164 	if (d->bd_filter)
   1165 		free((caddr_t)d->bd_filter, M_DEVBUF);
   1166 
   1167 	D_MARKFREE(d);
   1168 }
   1169 
   1170 /*
   1171  * Attach an interface to bpf.  driverp is a pointer to a (struct bpf_if *)
   1172  * in the driver's softc; dlt is the link layer type; hdrlen is the fixed
   1173  * size of the link header (variable length headers not yet supported).
   1174  */
   1175 void
   1176 bpfattach(driverp, ifp, dlt, hdrlen)
   1177 	caddr_t *driverp;
   1178 	struct ifnet *ifp;
   1179 	u_int dlt, hdrlen;
   1180 {
   1181 	struct bpf_if *bp;
   1182 	int i;
   1183 #if BSD < 199103
   1184 	static struct bpf_if bpf_ifs[NBPFILTER];
   1185 	static int bpfifno;
   1186 
   1187 	bp = (bpfifno < NBPFILTER) ? &bpf_ifs[bpfifno++] : 0;
   1188 #else
   1189 	bp = (struct bpf_if *)malloc(sizeof(*bp), M_DEVBUF, M_DONTWAIT);
   1190 #endif
   1191 	if (bp == 0)
   1192 		panic("bpfattach");
   1193 
   1194 	bp->bif_dlist = 0;
   1195 	bp->bif_driverp = (struct bpf_if **)driverp;
   1196 	bp->bif_ifp = ifp;
   1197 	bp->bif_dlt = dlt;
   1198 
   1199 	bp->bif_next = bpf_iflist;
   1200 	bpf_iflist = bp;
   1201 
   1202 	*bp->bif_driverp = 0;
   1203 
   1204 	/*
   1205 	 * Compute the length of the bpf header.  This is not necessarily
   1206 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
   1207 	 * that the network layer header begins on a longword boundary (for
   1208 	 * performance reasons and to alleviate alignment restrictions).
   1209 	 */
   1210 	bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
   1211 
   1212 	/*
   1213 	 * Mark all the descriptors free if this hasn't been done.
   1214 	 */
   1215 	if (!D_ISFREE(&bpf_dtab[0]))
   1216 		for (i = 0; i < NBPFILTER; ++i)
   1217 			D_MARKFREE(&bpf_dtab[i]);
   1218 
   1219 	printf("bpf: %s%d attached\n", ifp->if_name, ifp->if_unit);
   1220 }
   1221 
   1222 #if BSD >= 199103
   1223 /* XXX This routine belongs in net/if.c. */
   1224 /*
   1225  * Set/clear promiscuous mode on interface ifp based on the truth value
   1226  * of pswitch.  The calls are reference counted so that only the first
   1227  * "on" request actually has an effect, as does the final "off" request.
   1228  * Results are undefined if the "off" and "on" requests are not matched.
   1229  */
   1230 int
   1231 ifpromisc(ifp, pswitch)
   1232 	struct ifnet *ifp;
   1233 	int pswitch;
   1234 {
   1235 	struct ifreq ifr;
   1236 	/*
   1237 	 * If the device is not configured up, we cannot put it in
   1238 	 * promiscuous mode.
   1239 	 */
   1240 	if ((ifp->if_flags & IFF_UP) == 0)
   1241 		return (ENETDOWN);
   1242 
   1243 	if (pswitch) {
   1244 		if (ifp->if_pcount++ != 0)
   1245 			return (0);
   1246 		ifp->if_flags |= IFF_PROMISC;
   1247 	} else {
   1248 		if (--ifp->if_pcount > 0)
   1249 			return (0);
   1250 		ifp->if_flags &= ~IFF_PROMISC;
   1251 	}
   1252 	ifr.ifr_flags = ifp->if_flags;
   1253 	return ((*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr));
   1254 }
   1255 #endif
   1256 
   1257 #if BSD < 199103
   1258 /*
   1259  * Allocate some memory for bpf.  This is temporary SunOS support, and
   1260  * is admittedly a hack.
   1261  * If resources unavaiable, return 0.
   1262  */
   1263 static caddr_t
   1264 bpf_alloc(size, canwait)
   1265 	register int size;
   1266 	register int canwait;
   1267 {
   1268 	register struct mbuf *m;
   1269 
   1270 	if ((unsigned)size > (MCLBYTES-8))
   1271 		return 0;
   1272 
   1273 	MGET(m, canwait, MT_DATA);
   1274 	if (m == 0)
   1275 		return 0;
   1276 	if ((unsigned)size > (MLEN-8)) {
   1277 		MCLGET(m);
   1278 		if (m->m_len != MCLBYTES) {
   1279 			m_freem(m);
   1280 			return 0;
   1281 		}
   1282 	}
   1283 	*mtod(m, struct mbuf **) = m;
   1284 	return mtod(m, caddr_t) + 8;
   1285 }
   1286 #endif
   1287 #endif
   1288